The rapid evolution of medical imaging has shifted from conventional structural assessment toward a digitally driven paradigm in which Artificial Intelligence (AI) plays a central role in diagnostic workflows. While traditional modalities including Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and Positron Emission Tomography (PET) have enabled non-invasive diagnosis, they remain limited by challenges such as ionizing radiation exposure, inter-observer variability, and increasing global imaging workloads. This review synthesizes recent advancements in medical imaging AI from 2024 to 2026, with a focus on the transition from task-specific Convolutional Neural Networks (CNNs) to more generalizable architectures, including Vision Transformers (ViTs) and Medical Foundation Models. We examine the integration of Large Multi-modal Models (LMMs), which combine imaging data with clinical information such as electronic health records to enable more context-aware diagnostic reasoning. In addition, the role of generative AI in image reconstruction is discussed, particularly in relation to radiation dose reduction and accelerated image acquisition. By addressing the persistent gap between algorithmic development and real-world clinical deployment, this review highlights how emerging AI frameworks are contributing to safer, more efficient, and increasingly personalized diagnostic practices.
Joana Jan Kumi (Thu,) studied this question.